Temporis

Temporis Foundry is in closed beta

Train your own time series model on your own data.

Temporis Foundry is a platform for training and deploying generative time series models. Teams with a lot of private data use it to get a model that fits their domain.

Request access Read the Foundry docs
Your history Sampled futures event: "promotion starts"

A generative model samples many possible futures, and an event written as text can change them. Illustration, not real data.

From your data to a model in production

We bring the recipe for training a time series model. You bring the data. The model you get is yours.

  1. Bring your data

    Start from the time series your team already collects.

    Temporis Foundry

  2. Train your model

    Run our training recipe on your own data, on GPU machines that Foundry sets up for you.

    Temporis Foundry

  3. Deploy it

    Deploy the trained model from Foundry and start forecasting.

    Temporis Foundry

  4. Serve it online

    Keep the model fed with fresh data from a time series database, behind an API.

    Temporis Platform

What makes Temporis different

Google, Amazon, Nixtla and Datadog each have a time series foundation model of their own. Temporis lets you train yours.

Other time series foundation models Temporis
The model The vendor's model Your own modelTrained on your own data, so it fits your domain.
The forecast Typically not generative GenerativeThe model generates possible futures, one sample at a time.
Events No text event conditioning Event conditioningCondition a forecast on events written as text.

Temporis Platform, for serving your model online

A time series model in production needs a time series database. Platform is that database, built to sit on top of Foundry.

  • Time series databaseStores the data your model reads when it is serving online.
  • API basedSend data in and get forecasts out through an API.
  • MonitoringMonitoring for your deployment.
  • ManagedA managed system, running a custom model that fits your domain.
Open Temporis Platform Read the Platform docs
Temporis Platform overview screen with sections for access tokens, data sources, data profiles, API guide and billing

For anyone with a lot of private time series data

Companies and individuals who have the data and want a model of their own, in fields such as:

  • Quantitative trading
  • Energy
  • Traffic
  • Parking
  • E-commerce
  • Supply chain

Foundry is in use at UC Berkeley

The team

We come from:

  • Google
  • Meta
  • TikTok
  • CMU
  • UIUC

Want a time series model of your own?

Foundry is in closed beta. Email hello@temporis.co to ask for access, or to talk to us about anything else.

Request access Sign in to Foundry